IoT Analytics: At Edge or on Cloud?

Co-author: Ritesh Sinha

Introduction

Three crucial aspects of Big Data- velocity, variety, and volume (3Vs) realize their true potential in the Internet of Things (IoT). As a result, efficient data management (storage and processing) within the system is of utmost importance. It is only by applying advanced data processing and analytics tools that data’s fourth V- value can be derived.

In an IoT ecosystem, data is generated at various stages – devices & touch-points, the network they are connected to, the Edge systems forming the gateway, as well as the Cloud where the data is finally stored prior to application.

While device control data needs to be available either on the device itself or at the network level, additional data generated by the device can be processed ‘on the Edge’ or performed in the Cloud – leveraging the available virtual storage and computational capabilities.

So the question is,

Where do we perform majority of data computing tasks in IoT ecosystem – Edge systems or Cloud?

When it comes to creating an effective IoT ecosystem – Edge Analytics and Cloud Analytics – both have their own set of benefits and challenges.

Let us explore the various factors associated with the most popular data processing application - IoT analytics.

Criteria for Applying Analytics

An instance of analytics usage in IoT can be useful only when -

1. It assists in generating or solving business requirements and use cases
2. It is supported by hardware capabilities to perform the same
3. It is cost effective and efficient

The nature of a specific business use case determines the type of analytics applications that is written within that context. Let us consider some typical use cases that drive IoT analytics, and accordingly determine the location for analytics execution:

Use Cases

Case 1:After analytics is performed, if the result is used to control certain device parameters (prescriptive), then there are two factors that must be considered -

• If the control action needs to be relatively real time, then the action latency and efficiency are determining metrics – making an Edge device the preferred location.
• If the action is scheduled for a future timeline, or will be used for planning subsequent offline/online activities, then execution can be on Cloud and the result application can be deferred till required.

Similarly, if data computing in the above scenario requires higher CPU capacity – which Edge systems might lack – then they must be performed at the Cloud level. However, even then latency can prove to be a decisive factor.

Case 2:If the objective is to offer actionable insights on devices features, functionalities, and performance (descriptive), the following considerations should be made -

• If the data consolidated from the devices to the Cloud is voluminous, and the quality of results is not significantly compromised, then the Edge gateway can perform the aggregation/transformation process – reduce the size of data streams before sending them to Cloud.
• If the Edge gateway doesn’t possess adequate computing capabilities and/or if the quality of the results is getting affected, then the raw data must be sent to the Cloud without any intermediate processing.

Case 3:When analytics require data interpretation that incorporates local inputs, the following needs to be considered -

• If the data needs to be transformed with the local information before sending to Cloud, then the same has to be performed using Edge gateway.
• If the data just needs to be presented along with the local information for Cloud usage, then the Edge computing device needs to integrate local information with device data, and pass it on.

It should be noted that this application typically requires historical data to generate accurate results – consequently, the analytics process occurs on the Cloud.

Case 4:For applications in industries with disparate/heterogeneous data formats and networking environments – for instance, when data is generated and transmitted by devices in CAN or other industrial automation format, and via TCP/IP for Cloud communication – the determining factors include -

• If the data transformation requires interchange of standard types, then the Edge gateway proves useful.
• If the data can be transmitted without transformation and/or it is needed in the same format for future use, storage on the Cloud in that format is recommended.

There could be other use cases where similar decisions have to be taken – however, the examples above offer a general guideline and can help identify a pattern.

Other Conditions

There are other issues that need to be taken into account before a choice between Edge and Cloud is made. This includes gateway cost restrictions, available compute capabilities, storage capacity (both transient and persistent), and environmental factors surrounding gateway deployment, among others.

Security is another crucial aspect - businesses will not keep data where it is vulnerable to theft and fraud. Security is an important facet of IoT application which also contributes to the decision of where analytics should be performed.

Conclusion

As indicated above, IOT systems need to adopt a hybrid approach – performing Edge analytics and Cloud analytics based on current use cases. Further, they need to drive capabilities to support evolving features and future needs.

With the increasing maturity of Edge gateway systems, some of the decisions earlier made within an IoT ecosystem must be revisited. To drive maximum value from investments, it will be crucial for organizations to work with a team that can understand and deliver best-of-breed analytics performance.

References

1. The Evolution of IoT Edge Analytics: Strategies of Leading Players:http://www.kdnuggets.com/2016/09/evolution-iot-Edge-analytics.html